AI Workflow Designer & LLM Evaluator
Prompt engineering, LLM evaluation, and AI workflow design — with a focus on what actually works outside of demo environments.
I've been working with AI since 2022 — before it was a career, when it was just a tool that made my work sharper. What started as writing better English copy in a Web3 community evolved into designing full AI workflows, evaluating LLM outputs across five major models, and building systems that solve real problems for real people.
My background is unusual: 5+ years in crypto and Web3 gave me a sharp eye for hype versus substance — a skill that turns out to be extremely useful when evaluating AI. I know what a hallucination looks like. I know why a model fails in Russian. I know which model to use for which task, and I've built the prompt systems to prove it.
I work across Claude, GPT-4, Gemini, Perplexity, and NotebookLM — and I'm fluent in both English and Russian, which lets me catch the cross-lingual edge cases most evaluators miss.
Real capabilities, no exaggeration
I design multi-layered prompt systems using role prompting, chain-of-thought, context anchoring, few-shot calibration, and TCREI framework. Built 15+ production-level prompt systems for different platforms and use cases.
I test AI outputs against real-world use cases, identify failure patterns and edge cases, detect hallucinations, and spot multilingual degradation — especially in Russian-English cross-lingual contexts.
I build end-to-end AI workflows using Perplexity Spaces, NotebookLM, and no-code tools. Designed systems for legal document drafting, educational content generation, and research pipelines.
I build functional web products using Lovable, Replit, Cursor, and Vercel — without traditional engineering. Two live products deployed. GitHub for configuration and customization.
I write long-form AI-focused content in English and Russian — 12 published articles, 50k+ views across VC.ru, DTF.ru, and Dev.to. I research, structure, and publish independently.
I've systematically tested Claude, GPT-4, Gemini, Perplexity, and Grok across content generation, classification, structured reasoning, and workflow support — and documented what works when.
Real projects — personal experiments, systems I built, and content I published. No fake case studies.
Designed a Perplexity Space that answers legal questions based exclusively on Belarusian law. Used it to draft official complaints and responses to government bodies — saving lawyer fees and time. Basis for a planned Telegram bot product.
Built an AI-powered system in NotebookLM to generate curriculum-aligned educational materials — lessons, presentations, workbooks, and podcasts — based on approved Belarusian school programs. Designed for teachers, tutors, and parents.
Designed a suite of 15 production-level prompt systems — each tailored to a specific platform (LinkedIn, Threads, Twitter/X, VC.ru, Dev.to, Dzen) and use case. Includes role prompting, chain-of-thought, few-shot calibration, and TCREI framework.
Built and deployed a functional web application using AI-assisted development tools — no traditional engineering background. Live on Vercel.
Published 12 articles on AI tools and workflows across VC.ru, DTF.ru, and Dev.to — in Russian and English. 50,000+ total views, organic. Topics include AI video costs, LLM tool comparisons, and prompt engineering for non-technical users.
A sample of the systems I build — each one is a complete agent with defined role, context, style controls, and chain-of-thought logic.
Full persona system with TCREI framework, chain-of-thought decoding, visual cues, and style enforcement. Translates raw Russian thoughts into native-level English LinkedIn posts — maintaining authentic voice across platforms.
Context-anchored RAG system that constrains model responses strictly to Belarusian law sources. Designed for citizens navigating bureaucracy — produces legally accurate letters and complaint responses.
Multi-output prompt that generates article draft, 3 headline variants, positioning note, platform-specific tags, and social snippets — all maintaining the author's voice and avoiding AI-sounding language.
Five-level prompt engineering training system inside NotebookLM — from beginner to expert — with progressive exercises, feedback loops, and strict source grounding. Built for self-study in one intensive day.
Articles on AI tools, workflows, and practical applications — in English and Russian.
Total views
Articles published
Languages
Open to full-time roles, freelance projects, and consulting — in AI evaluation, prompt engineering, and workflow design.